Beliefs and Relief: The Impact of Political Uncertainty on

Small-Business Behavior in Uganda

Craig McIntosh*

December 2004

DRAFT

Abstract This paper studies the impact of the Ugandan presidential election of March 2001 on women running small businesses. Using data from the country’s largest microfinance lender along with opinion polls, county-level election results, and data from a newspaper survey, we study how promises of patronage and unexpected voting swings cause businesspeople to alter savings, investment, and expenditures. We find that this group of businesswomen believe that patronage exists in Ugandan politics, that not all talk is cheap, and that political events impose real economic shocks on the consumption in the households of entrepreneurs. Keywords: elections, patronage, uncertainty. JEL classification: O16, N27, H42

* International Relations/Pacific Studies, U.C. San Diego, 9500 Gilman Drive, La Jolla, CA, 92093-0519. ctmcintosh@ucsd.edu. Thanks to Stephen Haggard and Devra Coren Moehler, to Joanitah Kirunda, Robert Apell, and the staff of FINCA/Uganda, and to Mukalazi Deus and Jennifer Bartlett for excellent research assistance. All remaining errors are my own.

1

1. Introduction

The promotion of democratic elections has been a cornerstone of the policy prescribed

to Africa by Western nations over the past decade. While the long-run impact of

democratization is likely to be positive (Ndulu & O’Connell, 1999), evidence of the

disruptive effects of the elections themselves is growing. Whether triggered by competition

for popular support (Mansfield & Snyder, 1995), the instability brought about by the change

in leadership itself (Gray & McPherson 2001), or illiberal behavior by victors subsequent to

elections (Zakaria 1997), there is ample reason to think that electoral cycles generate

considerable uncertainty in the business environment. A growing body of literature uses

public opinion within developing countries1 (Bratton 2004) to understand what citizens

perceive are the risks and benefits of the electoral process, but this paper takes a different

approach. By studying changes in investment and consumption behavior among a group of

women running small, informal businesses through the election period, we are able to infer

what they believe and what is uncertain about the political system in which they operate.

The informal business sector is the mainstay non-agricultural economic activity in most

African economies, both in terms of income generation and the provision of goods and

services (Daniels & Mead, 1998). The way in which political expectations, uncertainty, and

patronage alter the behavior of such businesses is likely to have a major influence on the

flow of goods to the poor. This paper studies the business and household spending patterns

of the women who make up the clientele of FINCA Uganda, the country’s largest micro-

finance institution, during the presidential election. Through the results of two pre-election

opinion polls and the elections themselves, we can utilize spatial variation across counties

1 An excellent example is the Afrobarometer surveys, which now have two rounds of data available for Uganda.

2

and temporal variation within businesses to track how different sets of political events

manifest themselves in micro-level business response.

The 2001 election between incumbent Yoweri Museveni and the upstart Movement

insider Col. Kizza Besigye provides an interesting case study because of the absence of an

explicitly ethnic dimension2 and because it featured an opposition organization built from

the ground up in less than a year. In the absence of entrenched ethnic interests an electoral

competition opened up which was unusually issue-oriented, with a vote for the opposition

sending a message of change. The opposition scored well among the educated, urban,

young, and male, and Museveni’s core supporters were the older rural women with strong

memories of the country’s troubled past. This paper attempts to understand expectations

among businesswomen in the period around the election; the value of this approach is that,

in a political realm where perceptions may be more important than reality, it allows us to use

business people’s responses to events to infer what they believe to be political realities on the

ground.

2. Actions and Uncertainty.

Why would we expect to see behavior among micro-enterprises change in response to

political events? We suggest an approach to this problem which is derived from event

analyses in the finance literature (see, for example, Wilson et al 2000). In a world where

agents have access to credit and savings instruments (which FINCA’s clients obviously do),

we should see intertemporal investment being made on the basis of expectations of future

returns & liquidity needs. Thus, under the assumption that political events do not directly

2 Besigye is from the same ethnic group and region as the incumbent, served as Museveni’s personal doctor during the bush war in the 1980’s, and is married to Museveni’s ex-lover.

3

effect the access to financial instruments, any changes in behavior which result from political

shocks should reveal the extent which the revealed information alters expectations about the

future.

Rationally forecasting businesses will already have incorporated all predictable

information into their behavior. Thus, regardless of the exact form of political spending

prevalent in Uganda (see Magaloni, Estevez, & Diaz-Cayeros for evidence of patronage

flowing to closely contested districts in Mexico, and Weldon & Molinar (1994) for evidence

of support for core supporters), to the extent that businesswomen can forecast the voting

outcomes in their districts, they should already have built responses in the flow of patronage

into investment decisions. The key distinction here is that the election is only an event in the

informational sense if it yields a surprise outcome. Keefer (2003) among others has

suggested that less-developed democracies will feature clientelism (providing private goods)

more predominantly than pork-barrel spending (providing local public goods). Since we do

not observe patron-client relationships, or indeed individual voting, we cannot contribute to

this debate using our data. All that is implied by this story is that there is a sufficiently strong

general equilibrium effect from the flow of political spending that a link is created between

local voting outcomes and the local business climate.

We lay out here a series of loose hypotheses to help organize thoughts:

H1. Political events will have no impact on the trajectory investment behavior unless they

are both credible and unexpected.

H2. Any response to verbal threats or to promises of patronage imply that these statements

are both credible and unexpected, meaning that talk is not cheap.

H3. Any discontinuous change in outcomes in counties that voted along with expectations

implies that the outcome of national elections was uncertain to businesswomen.

4

H4. A significant coefficient on the size of the surprise vote for the winner (Movement)

implies a belief in patronage/punishment behavior by government among businesswomen.

H5. Shocks to sales in client businesses may filter through to all other behaviors; thus we

can only view changes in investment and consumption as ceteris paribus if sales remain

unchanged.

H6. Food consumption is the outcome which households have the strongest incentive to

save; consequently changes to food expenditures in response to political events indicates that

a shock has an impact which the household is unable to smooth away.

3. Data.

There are four distinct sources of data brought together in this paper. The first is data

from the client records of FINCA/Uganda, which includes detailed surveys on business

characteristics for a subset of clients. From these records, we have the following data: loans

are for a 16-week term, to be repaid weekly at an 87% effective annualized rate, and

fluctuations in borrowing can be seen as clients’ subjective short-term forecasts of the

business environment. We see savings in FINCA; these are physically held at Standard

Chartered Bank in group accounts (all FINCA lending is done in groups of 30) and these are

not demand deposits. As these savings are seized by FINCA in the event of failure to repay

and are usually only available at the end of the borrowing cycle, they are one of the less

liquid forms of saving available to these clients. We surveyed clients as to their daily sales in

their businesses over the past month, and the total value of all physical stock (including

working capital) they have invested in their business. Finally we asked clients to estimate the

total expenditures and the food expenditures per month in their households; the numbers

used here are per-capita figures.

5

Our second source of data is county-level election outcomes, including votes cast for

each of the candidates and the total vote. We use the percentage of the vote in each county

cast for the incumbent Movement candidacy of Yoweri Museveni and the total percentage

cast for the opposition candidates combined (Kizza Besigye, along with Moody Awori,

Chapaa Karuhanga, Kibirige Mayanja, and Francis Bwengye). We have accounting data for

FINCA clients located in 71 different counties.

Third, we have a newspaper survey which covers the months of November 2000

through July 2001. Uganda has two major newspapers; the New Vision, owned by the

Movement government, and the Monitor, owned by the Aga Khan’s Nation Media Group.

Only events which could be located at a specific place and time were included, because we

use time- and county-level fixed effects throughout, precluding the identification of country-

wide impacts.

The final source of data is a pair of opinion polls released by the New Vision on

February 3rd and February 28th 2001. These polls report outcomes only at the district level,

and results exist for only 19 of Uganda’s 56 districts. FINCA groups exist in 22 districts, of

which only 11 overlap with those covered by the opinion polls. From these we have

constructed a county-level prediction of voting based on that in found in the district as a

whole, or that in the closest district for which data exist. While this measure is very

imperfect, it also represents the best information available to voters in the run-up to the

election as to likely voting outcomes; no other polls existed, and it is not unreasonable to

think that voters were using polls to form expectations in a similar manner.

The FINCA data is a panel in which each individual features only ever four months

(or more) as they ‘recapitalize’ their loans. For this reason, it is difficult to interpret

discontinuous changes between time periods because they consist of different individuals.

6

7

The temporal sorting process is non-random, a fact which is easily verified by picking

‘placebo’ discontinuity dates, of which nearly 50% are significant. In order to minimize

differences between cohorts, we difference outcomes. We use only individuals for whom at

least two adjacent observations exist, using the two observations more distant from the

election to calculate a trend, and the dependant variable is the deviation from this predicted

outcome. Thus letting citY be a prediction formed from },,,{ 2112 ++−− citcitcitcit YYYY for the

outcome of individual i in county c and time t, we use the differenced outcome

citcitcit YYY −=&& to estimate how FINCA outcomes change over time across individuals.

The quality of the FINCA data varies. The savings and borrowing numbers, because

they pass through the accounting system and are double-booked, are very reliable and are

present for virtually all clients. The other individual-level statistics were collected in surveys

administered by FINCA’s credit officers at the time of receiving a new loan, and a great deal

of information is missing. Group-level data was collected in a separate survey, and again

suffers from missing observations. The use of the differencing process limits our number of

observations to individuals for whom we have three separate realizations of outcomes, and

corresponding control data as well. The numbers of observations in regressions using the

accounting data and only county-level controls is over 40,000. Using the survey-collected

outcomes, however, the number falls to about 5,000. When we run regressions where a

battery of individual-level controls must also be present, the number of observations falls

below 2,000 in some cases.

While this is sufficient data to obtain reasonably precise estimates, we are of course

concerned with the selection bias. An analysis of the observations for which we are missing

outcomes shows that the process is non-random; we are more likely to have data from rural,

older, ethnically homogeneous groups. Potentially more worrying, the groups for which we

8

have data are significantly more likely to vote for the Movement, and to experience violent

events. Throughout the analysis we use fixed-effects to remove group mean outcomes - all

identification comes from deviations from county-level averages, and so this problem is not

fatal. However, we must be aware that the group in which we estimate these effects (other

than for savings and loans) is neither the population nor necessarily a random sample of the

clients of FINCA.

4. Analysis of Correlations.

We now use these various data sources to explore several basic questions about the

timing and location of the events during the election period, and how business & household

behavior changed in response.

4.1 How did aggregate business behavior change during the election?

Figure 1 shows graphs of changes in the outcomes which we measure around the time of the

election. To calculate these graphs, we regress

cittw wr

rccit weektttY εβδδδδα ++++++= ∑∑=

4

1t

33

221 round &&

Where tround is a set of four dummies for each of the four-month rounds in which the

borrowers take out loans to drop out cohort-specific effects, and tweek is a set of week

dummies for w weeks around the election. Each week coefficient is therefore identified

relative to the (roughly) half of the data which is not within those w weeks. The vertical bar

in these graphs represents the day of the election, and moving out on the X axis gives us

variations in outcomes around that time. We control for a time trend, the squared and

cubed value of the time trend, and we report the point estimates and standard errors for wβ .

These graphs raise some interesting questions. First, we see savings and loan volumes

both decreasing shortly before the election, but they do not bottom out until more than a

month after the election, at which point they return quickly to normal. The decreased use of

formalized savings suggests that households are trying to hold more cash in hand; however

we do not see households running up precautionary debts in order to boost liquidity. Sales

remain roughly unchanged through the election period, with the exception of a slight

increase in activity 3-5 weeks after the election. Most intriguing, perhaps, is the dramatic

draw-down of business stocks in the weeks prior to the election, and the almost immediate

return to normal once the election has occurred. Indeed, the day of the election almost

perfectly picks up the turning point in business stocks. Total household expenditures, as

well as food expenditures, show a similar pattern to stocks in the weeks before the election,

but then take a somewhat puzzling plunge roughly a month after the election.

4.2 What is the nature and timing of the events from the newspaper survey?

Table 1 shows the frequency of each event, Table 2 the months in which the events

occurred, and Table 3 the number of events which occurred in each county. Movement

officials are roughly three times as likely to threaten the opposition as vice versa, and

physical attacks by the Movement and on opposition officials are similarly more prevalent

than those by the opposition and on the Movement. Promises of patronage, however, are

roughly symmetric from each side. Security was increased (usually implying the use of the

NRM military to keep the peace) 14 times, and election results were contested in 27

counties. The most prevalent political event from our survey is ‘Government threats to

opposing officials’, a category which runs the gamut from “Awori posters and campaign

materials missing from 5 districts” to “Security men storm Besigye’s office in Fort Portal

Town”. Nearly 80% of the time no events occur in a given month and county.

9

In Table 2 we see an acceleration of political instability leading up to the election in

March 2001, with only scattered incidents thereafter. 33% of the election-related events

occur in the same month as the election, and 90% in the months of Jan-March, 2001. Table

3 shows the concentration of events in urban areas; over 12% of the events occur in

Kampala City Council, the tiny county which contains the Parliament, many ministries, the

residence of the president, and the vast Owino market. The whole city of Kampala accounts

for almost 27% of the events, and the towns of Mbarara, Jinja, Arua, and Masaka account

for another 15%. The two more rural areas which account for a reasonable proportion of

the activity are Luwero, scene of the fiercest fighting of Uganda’s civil war and Rukungiri,

the home area of both Besigye and Museveni.

4.3 How do violence & civil unrest relate to voting outcomes?

Figure 2 illustrates a different take on the data by plotting the results of several

variables from the newspaper survey against county-level voting outcomes. Several

interesting regularities emerge from these correlations. In Figures 2.a.1. and 2.a.2, we see

how the incidence of patronage promises varies with the eventual vote in each county. The

Opposition promises patronage to safe seats, while the Movement makes promises across a

somewhat broader range. Political promises to ‘safe’ seats imply that political parties are

uncertain about county-level voting outcomes; or (because Uganda’s election is decided by

an absolute majority, not an electoral college) that they perceive higher marginal returns in

safe seats. The opposition may thus simply be mis-targeting its patronage promises, or it

may be that they gain more absolute votes by targeting promises to opposition strongholds.

Figures 2.b.1-4 illustrate the particularly tense nature of contested counties. 2.b.1

shows the number of violent acts per county, which is as we might expect concentrated in

contested counties, but more surprisingly carries over into Movement strongholds. Not a

10

11

single act of violence is recorded in counties where the opposition vote tops 70%. Similarly,

security was increased (usually entailing the use of soldiers to bolster extant police forces),

and government officials threatened, almost exclusively in contested counties. Attacks on

opposition officials cluster around a 50-50 vote, and again we see the surprising pattern that

such attacks take place almost exclusively in Movement-majority districts.

These patterns are reinforced by Figures 2.c.1-4; Movement strongholds and

contested counties are the source of all of the threats both to opposition and Movement

officials (the outlier in 2.c.1 is Kampala City Council, which saw more than its share of

political turbulence), as well as being the location where the election results themselves are

most likely to be contested. Threats against citizens, on the other hand, do not portray this

pattern, being situated almost exclusively in contested counties. The pattern of contestation

of election results in 2.c.4 is open to several interpretations; possibly it was only in

Movement districts that officials saw fit to challenge the tally; more likely the heavy tilt

towards the Movement in this figure is a direct product of rigging, which is widely believed

to have been conducted in the Movement’s favor. TP

3PT

5. Regression Analysis.

To test our hypotheses, then, we use the dependant variables described above to investigate

the following questions:

5.1 How does the election period differ for counties according to polling vs. voting?

We run the following:

citg

ctgmccit BY εβδα +++= ∑=

6

1

&&

TP

3PT Indeed, Rukungiri (Besigye’s home district) saw over 80% of the vote go to Museveni, and over 95% in

Kinkizi City. This county had previously been dubbed ‘vote rigging county’ (Monitor, Sept 2000) and is thought to have seen rigging on a large scale in the 2001 election.

12

where mδ is a set of month dummies, cα a county-level fixed effect, and ctB a set of

variables equal to -1 within a month before the election in group g, 1 within a month after,

and zero otherwise. The coefficients gβ thus measure the average swing in outcomes from

pre- to post-election, comparing these changes to the time-detrended within-county

variation. The following taxonomy of county-level outcomes is exhaustive:

Group 1: Polls predict Movement victory by more than 60%, Movement wins county. Group 2: Polls predict a Movement victory by more than 60%, opposition wins county. Group 3: Polls predict an opposition victory by more than 60%, Movement wins county. Group 4: Polls predict an opposition victory by more than 60%, opposition wins county. Group 5: Polls are between 40%-60%, Movement wins county. Group 6: Polls are between 40%-60%, opposition wins county. Groups 1 & 3, then, are safe seats in which no county-level surprise occurs, groups 2

& 4 show the largest county-level surprise, and groups 5 & 6 are contested counties. Table 4

gives the coefficients gβ for the six groups in g, and for the six different business-level

outcomes from FINCA clients.

First, we note that the lack of impacts on sales in Table 4 means that other business

variables can be viewed as household liquidity and consumption decisions, as the demand

side of client businesses is ceteris paribus. Thus any discontinuous change in outcomes in

counties that voted along with expectations implies that the outcome of national elections

was uncertain to businesswomen; otherwise there is no ‘event’. In the end, safe opposition

counties that go with the opposition display the strongest response. Here, local-level

outcomes did not differ from expectations, and yet there is a strong business response. In

these counties the primary information is the outcome of the presidential election itself,

showing that uncertainty over the outcome of the national election was real. The fact that

stronghold opposition districts show a strong response implies that they feel or anticipate a

bigger change in fortune than the contested districts.

13

The fall in expenditures in opposition strongholds is consistent with the overall

depressive effect of the Movement victory; more surprising is the similar fall seen in

Movement strongholds and contested counties which go to the Movement. A candidate

explanation for this is a flow of pre-election largesse which drove up consumption prior to

the election and dried up once the votes were in. None of these effects change food

consumption, implying that this kind of information shock can be fully smoothed.

One problem with this analysis is that all opposition candidates are lumped together

into one, whereas in fact they ran as individuals. Thus what appear to be ‘safe’ seats by the

aggregate opposition vote may not in fact be so. Besigye’s 27.7% of the vote, however,

represented over 90% of the total vote for the opposition, and so this is likely to be a

reasonable proxy.

5.2 What is the response to political shocks?

Here we estimate

cits

ctsmccit SY εβδα +++= ∑=

7

1

&&

where ctS is a set of seven dummies equal to one in the month after new political events are

revealed to have occurred in a county. Given the fixed effects, these dummies measure how

outcomes in the month after each kind of shock differ from the predicted county-level

detrended outcomes. Table 5 reports on sβ , and the political shocks are the following:

1. Acts of violence (any attacks, beatings, shootings, etc. which are politically motivated). 2. Threats against the opposition, which includes the following categories from the newspaper

survey: Government threats to opposition candidates, Government Arrests opposing supporters, Government threats to opposing officials, Government supporters threaten opposition, and Physical attack on opposition officials/supporters.

3. Threats against the Movement, includes Opponent supporters threaten government officials, Citizens threaten government, and Physical attack on government official/supporters.

4. Threats against citizens, includes Opponent supporters threaten citizens, Government/police attack/arrest citizens, Citizens struggle with citizens, and Increased security (included because it was an implied threat, and a response to violence).

5. Election results contested. 6. Promise of patronage made to a county by a politician of any party. 7. The size of the surprise vote for the movement; this variable takes a value of zero before the

election and one month after; for the month following the election it is (actual county-level percent vote for movement - predicted vote).

Given the strong effects of some of these shocks on sales, it is entirely possible that all

other effects seen are simply the result of non-separable household enterprises responding to

demand-side shocks. However, given the symmetry shown by the effects of threats to the

Movement and threats to the opposition, the implication is that such threats are viewed

similarly by the business community. An exception is that threats to Movement trigger more

food stockpiling. Threats against citizens have weaker effects but trigger the strongest food

stockpiling effect.

Acts of violence trigger an increased demand for liquidity and stockpiling in household

consumption. What seems counterintuitive is that violence causes these street-level

businesswomen to increase investment in inventories as well. Given the strong jump in own

consumption and the weak increase in sales, it may be that this is simply a business response

to demand-side changes induced by the violence. Inventories are clearly forward-looking as

they move in the same direction as the movement in sales; if they were myopic we would see

them being drawn down and hence would move in the opposite direction as sales.

The size of the surprise vote for the movement, another form of informational shock,

triggers the effects that we would expect given the evidence in 5.1; a boom in the use of

formal savings and borrowing when the vote moves in favor of the movement. In principle

this coefficient tests for the presence of the belief among Ugandan businesswomen that

county-level voting will alter future economic outcomes (H4). However, the possibility of

reverse causality here is clear; it may be that it is precisely the feeling of wellbeing evidenced

14

15

by the increase in formal-sector activity which makes the vote swing to the side of the

establishment.

Promises of patronage do in fact have a weak effect on increasing loans.TP

4PT While these

promises have no other effects, this is some evidence that not all talk is cheap (see Ferrel &

Rabin, 1996), as for there to be any effect at all these promises must be both credible and

unexpected. The fact that contestation of election results has no impact whatsoever implies

that businesswomen had no faith that political realities would change. An interesting

expectations story can be told; given the Movement victory and the predominance of

charges of rigging against the Movement, FINCA clients might reasonably have expected

that the actual vote in each county could be inferred by the victors, making the recounted

totals redundant to future plans for pork.

5.3 How do individuals differ in their response to political violence?

This analysis estimates

citcitctcitctmccit XVXVY εβγγδα +++++= )*(0&&

where ctV is a dummy indicating an act of political violence in county c within the past

month, citX vector of individual-level characteristics, and β measures the interaction.

From this analysis we see how the impact of this shock differs across individual

characteristics, as compared to differences from the county-level prediction of outcomes

where there was no shock.

Table 6 shows no difference across individual characteristics in the way that

loans, savings, expenditures, and sales respond to violence. That individual characteristics

TP

4PT Most localized patronage promises entail the tarmacking of roads, or the provision of water or electricity to

remote areas.

have no effects on sales is not surprising, but in the other variables is more so. The

differences come in the way that cash flow and food expenditures are managed. Looking at

business stocks, class indicators such as education and land ownership show that wealthier

sustain shocks better. Owning a home, however, increases the amount pulled out of stocks

when violence occurs; this may imply a safe place to which clients can pull out their stock.

Surprisingly, groups that pre-existed FINCA has bigger negative effect, contradicting better

mutual insurance in groups with better social capital.

Theory tells us that food expenditures should provide our best test of

consumption smoothing under shocks. However, we see the educated showing bigger

decreases in food expenditures and hence displaying less consumption smoothing. More

adults in the household lead to less smoothing; either consumption in these households is

less elastic, or they have a better ability to substitute own production for cash expenses.

Groups which conduct other informal savings arrangements among members smooth better.

Ethnically homogeneous groups smooth less well, consistent with the idea that diversity

improves the ability to mutually insure (although this effect could also show that ethnically

homogeneous groups are more exposed to political risk, as we would expect from an

extension of the argument in Miguel & Gugerty (2004): because homogeneous regions are

better at providing local public goods, they are more vulnerable to their loss).

A set of regressions which were run but are not reported here decompose effects

across the type of business operated by individual clients, and across the language spoken in

FINCA group meetings (e.g., ethnic group). Interestingly, these regressions show no

differentiation whatsoever in the response to political shocks across these categories. The

former result is very surprising, as it was hypothesized that street-level retailers shouls

demonstrate a much more intense response to acts of violence than, say, farmers. The latter

16

result confirms the non-ethnic nature of the election, but is also surprising. It may be that

the county-level fixed effects subsume most of the differential impacts of ethnicity, but

business activity varies widely within the same region and so should be fully identified by the

setup of the regression.

6. Conclusion.

These data provide strong evidence for the fact that women running small enterprises in

Uganda believe that their government is clientelistic, or will provide patronage. Their

behavior is more consistent with a scenario in which rewards flow to areas which provide

strong support (and away from stronghold opposition counties) than with one in which

patronage is focused on an intense competition for marginal counties. Disproportionate

promises of patronage by the opposition to safe counties suggests the inexperience of a

candidacy built from the ground up in several months because of Uganda’s ‘no-party’

system. The entrenched Movement, on the other hand, was more effective at targeting

patronage promises towards swing counties. An absence of strong ethnic identification in

the data, either in the diversity of borrowing groups or the language spoken in these groups,

suggests that the presidential election of 2001 did indeed transcend ethnicity.

Shocks caused by the electoral cycle have strong impacts on business activity. While

direct violence leads both FINCA clients and their customers to stockpile in consumption,

threats against either party caused a strong depression of sales, leading to adverse effects on

small enterprises. Ethnically diverse borrowing groups are better at smoothing consumption

in the face of such shocks.

Talk is not cheap in Ugandan politics. Not only do patronage promises have some

impact on outcomes (in essence, a muted version of the impact seen when voting outcomes

lead entrepreneurs to expect greater pork), but threats have strong general equilibrium

17

effects, demonstrating that FINCA clients and their patrons find them credible. The fact

that contestation of election results causes no local response, on the other hand, implies that

businesswomen do not believe that their economic prospects will be materially changed by

such challenges.

Through the use of revealed beliefs about the impact of political activity on the

economy, we gain insight into the beliefs and uncertainties in the minds of micro-

entrepreneurs. Clearly, the outcome of the election was in doubt in the minds of the voters,

and the end of the electoral cycle causes a gradual return to normalcy. While the election of

2001 featured an unusual focus on competition over issues, we see evidence that pork and

electoral largesse are largely focused on each party’s home counties, as we would expect in a

more ethnically-driven election. The uncertainties brought about by the campaign have

strongeffects at the household level, and induce shocks to consumption of food from which

the households are unable to protect themselves. Whatever the long-term benefits of the

democratic process, elections themselves impose real costs on citizens.

18

References

Bratton, M., (2004). ‘State-building and Democratization in Sub-Saharan Africa: Forwards, Backwards, or Together?’. Afrobarometer Working Paper No. 43.

Diaz-Cayeros, A., and Beatriz Magaloni (2003). ‘The politics of public spending. Part 1 – The logic

of vote buying.’ World Bank Background Paper. Diaz-Cayeros, A, Beatriz Magaloni and Federico Estevez (forthcoming) “The Erosion of Party

Hegemony, Clientelism and Portfolio Diversification: The Programa Nacional de Solidaridad (PRONASOL) in Mexico” in Herbert Kitschelt (ed.).

Gray, C. and M. McPherson, (2001). ‘The Leadership Factor in African Policy Reform and

Growth.’ Economic Development and Cultural Change, Vol. 49 No. 4, pp. 707-740. Daniels, L. & D. Mead, (1998). ‘The Contribution of Small Enterprises to National and Household

Income in Kenya.’ Economic Development and Cultural Change, Vol. 47 No. 1, pp. 45-71. Ferrell, Joseph, and Matt Rabin (1997). ‘Cheap Talk.’ Journal of Economic Perspectives, Vol. 10, No. 3,

pp. 103-118. Keefer, P. (2003). ‘What does political economy tell us about economic development – and vice

versa?’. World Bank Working Paper. Mansfield, E., and Jack Snyder (1995). ‘Democratization and War’. Foreign Affairs, Vol. 74 No. 3,

pp. 79-97. Miguel, E. and Mary Kay Gugerty, (2004). ‘Ethnic Diversity, Social Sanctions, and Public Goods in

Kenya’, working paper. Molinar-Horcasitas, Juan and Jeffrey A. Weldon (1994). ‘Electoral Determinants and Consequences

of National Solidarity.’ In Transforming State-Society Relations in Mexico: The National Solidarity Strategy, ed. Wayne A. Cornelius, Ann L. Craig, and Jonathan Fox. La Jolla: Center for U.S.-Mexican Studies, University of California, San Diego

Ndulu, B. and S. O’Connell, (1999). ‘Governance and Growth in Sub-Saharan Africa.’ Journal of

Economic Perspectives, Vol. 13 No. 3, pp. 41-66. Wilson, B., A. Saunders, and G. Caprio, (2000). ‘Financial Fragility and Mexico’s 1994 Peso Crisis:

An Event-Window Analysis of Market-Valuation Effects.’ Journal of Money, Credit and Banking. Vol. 32 No. 3, pp. 450-468.

Zakaria, F, (1997). ‘The Rise of Illiberal Democracy.’ Foreign Affairs, Nov/Dec., pp. 22-43.

19

20

Figure 1. Loan Volume

-60

-50

-40

-30

-20

-10

0

10

20

30

40

-10 -5 0 5 10

Weeks from election

Thou

sand

Ush

Coef.LowerUpper

Savings Volume

-15

-10

-5

0

5

10

-10 -5 0 5 10

Weeks from election

Thou

sand

Ush

Series1Series2Series3

Total Business Stock

0

50

100

-10 -5 0 5 10

h

-200

-150

-100

-50

Weeks from election

Thou

sand

Us

Coef.LowerUpper

Daily Sales Volume

15

20

25

30

0 5 10

s from election

sh

-15

-10

-5

0

5

10

-10 -5

Week

Thou

sand

U

Coef.LowerUpper

Total Household Expenditures

-5

-4

-3

-2

-1

0

1

2

-10 -5 0 5 10

Thou

sand

Ush

-7

-6

Weeks from election Coef.LowerUpper

Household Food Expenditures

-1.5

-1

-0.5

0

0.5

-10 -5 0 5 10

s from election

Thou

sand

Ush

-2Week Coef.

LowerUpper

Figure 2. Scatterplots of county-level voting outcomes versus various political events.

0.5

11.

52

# of

Pat

rona

ge p

rom

ises

by

Opp

ositi

on

0 20 40 60 80 100Percent vote for Movement

bandw idth = .8

2.a.1. Opposition patronage promises only in strongholds

0.5

11.

52

# of

Pat

rona

ge p

rom

ises

by

Gov

ernm

ent

0 20 40 60 80 100Percent vote for Movement

bandw idth = .8

2.a.2. Movement patronage promises evenly distributed

21

-20

24

6N

umbe

r of v

iole

nt e

vent

s in

cou

nty

0 20 40 60 80 100Percent vote for Movement

bandw idth = .8

2.b.1. No violence in Opposition strongholds

01

23

4#

of In

cide

nts

of in

crea

sed

secu

rity

0 20 40 60 80 100Absolute value of voting margin

bandw idth = .8

2.b.2. Security increased only in contested counties0

12

3#

of th

reat

s ag

ains

t Gov

ernm

ent o

ffici

als

0 20 40 60 80 100Percent vote for Movement

bandw idth = .8

2.b.3. Threats to government in contested counties

-10

12

3#

of a

ttack

s on

Opp

ositi

on o

ffici

als

0 20 40 60 80 100Percent vote for Movement

bandw idth = .8

2.b.4. Attacks on Opposition in contested counties

22

23

05

1015

2025

Tota

l # o

f thr

eats

aga

inst

Opp

ositi

on

0 20 40 60 80 100Percent vote for Movement

bandw idth = .8

2.c.1. Threats to Opposition from Movement Strongholds

01

23

45

Tota

l # o

f thr

eats

aga

inst

Gov

ernm

ent

0 20 40 60 80 100Percent vote for Movement

bandw idth = .8

2.c.2. Threats to Movement not in Opposition strongholds

02

46

Tota

l # o

f thr

eats

aga

inst

citi

zens

0 20 40 60 80 100Percent vote for Movement

bandw idth = .8

2.c.3. Threats to citizens from contested counties

01

23

45

Elec

tion

resu

lts c

onte

sted

0 20 40 60 80 100Percent vote for Movement

bandw idth = .8

2.c.4. Elections not contested in Opposition strongholds

Table 1. Frequency of Political Events, by Event (Unit: county/month).Event: Frequency Percent

0 No Event 963 78.481 Government Threats to Opposition Candidates 16 1.32 Opponent Supporters Threatens Citizens 2 0.163 Opponent Supporters Threaten Government Officials 11 0.94 Government Arrests Opposing Supporters 27 2.25 Government Threats to Opposing Officials 53 4.326 Government / Police Attack / Arrest Citizens 2 0.167 Peaceful Demonstrations 4 0.338 Citizens Struggle with Citizens 4 0.339 Promises of Patronage by Government 10 0.81

10 Promises of Patronage by Opposition 12 0.9811 Government Supporters Threaten Opposition 4 0.3312 Large-scale defections to Movement 2 0.1613 Citizens Threaten Government 5 0.4114 Physical Attack on Government Official/supporters 15 1.2215 Physicial Attack on Opposition Official/supporters 39 3.1816 Physical Attack by Government Official/supporters 8 0.6517 Physical Attack by Opposition Official/supporters 1 0.0818 Increased security 14 1.1419 Election results contested 27 2.220 Government arrests/threatens Government Official/supporters 8 0.65

Table 2. Frequency by Monthmonth Frequency Percent

Nov. 2000 2 0.57Dec. 2000 22 6.25Jan. 2001 103 29.26Feb. 2001 95 26.99Mar. 2001 116 32.95Apr. 2001 9 2.56May. 2001 4 1.14Jul. 2001 1 0.28

Total 352 100

24

25

Table 3. Number of events by county:District County # Events FINCA Operates? District County # Events FINCA Operates?APAC KOLE 2 yes LIRA DOKOLO 0 yesAPAC KWANIA 1 yes LIRA ERUTE NORTH 1 yesAPAC MARUZI 1 yes LIRA ERUTE SOUTH 2 yesAPAC OYAM 1 yes LIRA LIRA MUNICIPALITY 3 yesARUA ARUA MUNICIPALITY 8 yes LIRA MOROTO 2 yesARUA AYIVU 2 yes LUWERO BAMUNANIKA 2 yesARUA TEREGO 2 yes LUWERO KATIKAMU NORTH 4 yesARUA VURRA 2 yes LUWERO KATIKAMU SOUTH 9 yesBUGIRI BUKOLI 4 yes LUWERO NAKASEKE 1 yesBUGIRI BUKOLI SOUTH 3 yes MASAKA BUKOMANSIMBI 3 yesBUSIA BUSIA 5 yes MASAKA BUKOTO 7 yesGULU ASWA 3 no MASAKA KALUNGU 5 yesGULU GULU MUNICIPAL COUNCIL 2 no MASAKA MASAKA MUNI 6 yesHOIMA BUGAHYA 4 yes MASINDI BULIISA 0 yesIGANGA BUGWERI 1 yes MASINDI BURULI 3 yesIGANGA BUNYA EAST 1 yes MASINDI KIBANDA 0 yesIGANGA BUNYA SOUTH 1 yes MAYUGE BUNYARUTA 1 noIGANGA BUNYA WEST 1 yes MBALE MBALE 6 noIGANGA BUSIKI 1 yes MBALE MBALE MUNI 2 no

yesnoyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesnonononononoyesyesyes

IGANGA KIGULU NORTH 3 yes MBARARA NYABUSHOZI 13IGANGA KIGULU SOUTH 3 yes MOROTO MOROTO MUNICIPALITY 1IGANGA LUUKA 1 yes MPIGI BUTAMBALA 0JINJA BUTEMBE 2 yes MPIGI ENTEBBE 6JINJA JINJA MUNIC EAST 12 yes MPIGI MAWAKOTA NORTH 2JINJA JINJA MUNIC WEST 9 yes MPIGI MAWAKOTA SOUTH 1JINJA KAGOMA 1 yes MUBENDE BUSUJJU 1KABALE KABALE 2 no MUBENDE MITYANA SOUTH 1KABAROLE FORT PORTAL 2 no MUKONO BUIKWE NORTH 0KAMPALA KAMPALA CC 40 yes MUKONO BUIKWE SOUTH 1KAMPALA KLA KAWEMPE NORTH 9 yes MUKONO BUIKWE WEST 0KAMPALA KLA KAWEMPE SOUTH 9 yes MUKONO MUKONO NORTH 9KAMPALA KLA MAKINDYE EAST 9 yes MUKONO MUKONO SOUTH 4KAMPALA KLA MAKINDYE WEST 8 yes MUKONO NAKIFUMA 0KAMPALA KLA RUBAGA NORTH 2 yes NAKASONGOLA NAKASONGOLA 3KAMPALA KLA RUBAGA SOUTH 3 yes RAKAI KABULA 0KAMPALA NAKAWA 8 yes RAKAI KOOKI 0KAMULI BUDIOPE 0 yes RAKAI KYOTERA 1KAMULI BUGABULA NORTH 3 yes RUKUNGIRI KINKINZI CTY EAST 6KAMULI BUGABULA SOUTH 2 yes RUKUNGIRI KINKINZI CTY WEST 3KAMULI BULAMOGI 2 yes RUKUNGIRI RUKUNGIRI TOWN 7KAMULI BUZAYA 0 yes SOROTI SOROTI MUNICIPALITY 2KAMWENGE KIBALE COUNTY 5 no SSEMBABULE MAWOGOLA COUNTY 2KAYUNGA NTENJERU NORTH 4 yes TORORO TORORO 8KIBAALE BUYAGA COUNTY 1 no WAKISO BUSIRO EAST 1KIBOGA KIBOGA EAST 0 yes WAKISO BUSIRO SOUTH 1KIBOGA KIBOGA WEST 0 yes WAKISO KYADONDO 2KUMI KUMI TOWN 4 no

Table 4. Swings from the month before to the month after the election, conditional on polls and actual electoral outcomes:

Poll prediction, outcome: Loans Saving Stock Expenditures Food Exp. SalesCoeff. T-stat Coeff. T-stat Coeff. T-stat Coeff. T-stat Coeff. T-stat Coeff. T-stat

Movement, Movement wins 8.36 1.14 -0.61 -0.18 -246.27 -1.88 -3.21 -3.12 -0.47 -1.38 0.96 0.15Movement, Movement loses -15.06 -1.92 -4.85 -1.37 -3.82 -0.02 -0.33 -0.16 0.18 0.28 -0.28 -0.02Opposition, Opposition wins -210.75 -11.64 -28.10 -3.43 -321.47 -1.07 -7.11 -4.27 -0.12 -0.24 5.83 0.56Opposition, Opposition loses 0.29 0.04 4.02 1.16 -22.14 -0.2 -0.88 -1.03 0.07 0.28 1.24 0.23Contested, goes to Movement 2.94 0.38 2.38 0.68 -214.18 -1.42 -2.36 -1.97 -0.40 -1.06 5.52 0.73Contested, goes to Opposition -13.18 -0.74 2.68 0.33 95.05 0.26 (dropped) 0.49 0.51 1.37 0.07

# obs 40521 40514 4654 5253 4983 5461

Table 5. Event Study of the impact of information revelation on business behavior:Event: Loans Saving Stock Expenditures Food Exp. Sales

Coeff. T-stat Coeff. T-stat Coeff. T-stat Coeff. T-stat Coeff. T-stat Coeff. T-statAct of Violence -26.51 -3.14 -5.72 -1.46 309.26 2.94 3.80 4.66 0.63 2.4 7.89 1.55Threat against Opposition 2.85 0.4 -0.53 -0.16 -439.42 -4.63 -2.44 -3.22 -0.15 -0.64 -11.18 -2.36Threat against Movement 24.69 1.99 9.71 1.69 -349.24 -2.15 -6.35 -5.61 0.66 1.85 -16.99 -2.37Threat against citizens 26.38 1.83 -3.44 -0.51 337.35 0.87 2.91 0.91 2.61 2.67 35.05 1.74Election results contested 5.54 0.47 1.65 0.3 -212.80 -0.85 0.03 0.02 -0.12 -0.19 5.29 0.41Patronage Promise 29.75 2.01 -6.98 -1.02 -281.82 -0.24 -2.35 -0.24 -1.97 -0.66 -6.85 -0.11Surprise vote for Movement 0.58 3.51 0.23 2.96 0.49 0.25 0.00 0 0.00 0.04 0.04 0.48

# obs 40524 40517 4654 5253 4983 5461

26

Table 6. Decomposition of the impact of violence across individual & group characteristics:Characteristic, interaction: Loans Saving Stock Expenditures Food Exp. Sales

Coeff. T-stat Coeff. T-stat Coeff. T-stat Coeff. T-stat Coeff. T-stat Coeff. T-stat# children in HH -6.26 -1.28 -0.77 -0.28 -21.18 -0.3 -0.27 -0.53 0.27 1.6 4.18 1.58# adults in HH 10.40 1.52 4.33 1.14 39.84 0.31 -1.03 -1.56 -0.75 -3.41 -0.09 -0.02Client education -9.84 -0.76 -1.43 -0.2 449.56 2.03 0.95 0.78 -1.22 -2.91 -0.88 -0.13HH owns home -18.68 -0.76 -17.26 -1.27 -974.04 -2.25 1.73 0.58 -0.42 -0.4 5.45 0.34HH owns land 21.88 0.99 17.83 1.44 1070.73 2.41 -0.69 -0.23 0.84 0.79 16.37 1.01Client DOB 0.00 -0.72 0.00 -0.15 0.08 1.6 0.00 -0.77 0.00 0.11 0.00 -0.54Ethnic Heterogeneity of group 0.93 0.07 5.97 0.79 -388.11 -1.1 -1.64 -0.71 0.76 2.01 -8.12 -0.67Group Pre-existed FINCA 9.24 0.33 -7.70 -0.5 -1690.20 -2.97 -3.61 -0.8 -0.09 -0.05 -9.93 -0.41Group conducst ROSCA 7.08 0.29 2.37 0.17 -988.02 -0.78 1.86 0.58 3.19 3.09 6.37 0.37

# obs 17052 17052 1583 1993 1935 2060

27